The Facebook Ads audit checklist for high-spend accounts.

A Meta account at scale fails in different places than a search account, and the difference changes what an audit is for. On Google, most findings are waste — spend going somewhere it should not, which you exclude and redirect. On Meta there are fewer knobs to misconfigure and more ways to starve the delivery system, so most findings are capacity: something is capping what the whole account can achieve, and fixing it raises the return on every dollar rather than reallocating a slice of them.

That distinction matters at the end, when you have to tell someone what the audit is worth. It also explains why an account can look immaculate at campaign level and still be stuck: nothing is being wasted, the machine is simply being fed badly. Below is the checklist with the numbers attached, ordered from the signal the platform learns on outward to the creative it spends against.

1. Signal quality and the Conversions API

• →Confirm the Conversions API is live and healthy, not the browser pixel alone. Finding if absent: server-side signal is the difference between a learning account and a guessing one, and no campaign work compensates for it.
• →Check event match quality on hashed email, phone and other identifiers. Finding if match quality is low, and note the mechanism — poor matching does not degrade reporting gracefully, it removes conversions from the training set. The account is then optimizing on a biased sample of its own results.
• →Verify deduplication between pixel and CAPI events. Finding if dedup is unconfigured, regardless of whether totals look plausible — see the trap below.
• →Confirm purchase values are passed accurately for value optimization. Finding if value variance is near zero on a business with variable order values. A flat placeholder makes value optimization identical to conversion-count optimization, with extra steps.
• →For catalog advertisers: verify the feed is complete, current, and matched to pixel content IDs. Finding if content ID mismatch is present, because it silently breaks dynamic retargeting while everything continues to report normally.

2. Structure: the arithmetic that decides ad-set count

Meta needs roughly 50 optimization events per ad set per week to exit the learning phase and stay out of it. That number is not a guideline you can argue with by having better creative — it is a floor set by how the delivery system learns. And it means your total conversion volume, not your strategy, determines the maximum number of ad sets you can support:

Maximum learning-grade ad sets = weekly conversions / 50, rounded DOWN

400 purchases/month  ->  ~92/week   ->  1 ad set
1,000/month          ->  ~231/week  ->  4 ad sets
2,500/month          ->  ~577/week  ->  11 ad sets

An account at 400/month running 12 ad sets gives each one ~8
weekly conversions: every ad set permanently in learning, none
of them ever judged on stable delivery.

This is the most common structural finding at any spend level, and it is arithmetic rather than opinion. Run the division before looking at anything else in the structure, because if the account fails here then every ad-set-level performance comparison in it is a comparison between unstable estimates — including the ones somebody has been making budget decisions from.

• →Compute the ceiling above and compare it to the live ad-set count. Finding if the count exceeds the ceiling. The fix is consolidation, and it is usually resisted because it feels like losing control.
• →Confirm splits follow genuine business economics — margin, market, objective — rather than targeting hypotheses. Finding if any split exists to test an audience theory the delivery system tests better internally, because that split costs you learning-grade volume to buy information you were already getting.
• →Check audience overlap between ad sets bidding for the same people. Finding on material overlap, which is self-competition you are paying the auction premium for.
• →Verify each ad set holds enough budget to reach the 50-event floor at its actual cost per event. Finding if the budget arithmetic makes the floor unreachable — an ad set that mathematically cannot exit learning should not exist.

3. Creative volume and fatigue

• →Assess creative velocity: a real pipeline of net-new concepts, or three tired assets carrying the account. Finding if net-new concepts are not arriving on a cadence, because on Meta creative is the targeting layer and volume is a structural input rather than a nice-to-have.
• →Read fatigue as a decay rather than a level. Finding when performance has fallen materially from an asset's own established baseline and stayed there — roughly a quarter down, sustained. A single bad week is variance, and treating it as fatigue retires assets that were about to recover.
• →Hold a minimum-impressions floor before labeling anything. Finding is not available below a few hundred impressions in the window, no matter how bad the numbers look — the label is worth watching, not confirmed. The full decision procedure is in the creative testing system, which explains why ranking beats significance testing here.
• →Verify format and hook variety: static, video, UGC-style, different angles for different segments. Finding if the library is five variations of one idea, which gives the delivery system nothing to differentiate between and produces the appearance of testing without the substance.

4. Advantage+ and automation discipline

• →Check how Advantage+ campaigns are being judged. Finding if the decision rests on the ROAS they report about themselves rather than on whether total blended efficiency moved. An automated campaign that claims conversions from elsewhere in your mix reports beautifully while changing nothing.
• →Confirm new-customer settings or audience definitions where acquisition is the objective. Finding if absent on an acquisition account, because the automation will otherwise rebuy your existing customers and correctly report it as revenue.
• →Verify automation is fed genuine creative variety. Finding if the asset pool is thin — automation cannot rescue creative, it can only distribute it faster.
• →Look for full automation with no measurement scaffold. Finding if there is no blended read and no holdout, because you then have no way to distinguish delivering from claiming, which is the one question automation makes harder to answer.

5. Measurement honesty

• →Reconcile Meta-reported revenue against back-end revenue for the same window. Finding on a material gap, with the caveat from the callout above — check the two mechanisms separately rather than trusting a balanced total.
• →Confirm a blended metric is tracked and governs. Finding if platform-reported ROAS is the number budget decisions are made from — the reasoning is in MER versus ROAS, and the short version is that platform ROAS answers a question about the platform, not about the business.
• →Check the attribution window and whether anyone has changed it. Finding if the window was widened at any point that coincides with a reported improvement, which is a reporting change wearing the costume of a performance change.
• →Separate new-customer from returning revenue. Finding if the split does not exist, because acquisition efficiency is then unknowable and a growing repeat base will make acquisition look like it is improving while it deteriorates.

What the findings are worth

Here is where a Meta audit needs different arithmetic from a Google one. A waste finding is sized by the spend it redirects: exclude $12,000 of genuinely wasted budget and, at a 4.0 target, you have found roughly $48,000 of incremental revenue rather than $12,000 of savings. That logic applies cleanly to search waste, and barely applies here.

Meta findings are mostly capacity, and capacity findings apply to the whole budget rather than a slice of it. Fixing degraded signal does not redirect spend — it improves the quality of every delivery decision the system makes with all of it. Consolidating eleven ad sets into four does not save money; it moves the account from eleven unstable estimates to four reliable ones, which changes what every future decision is based on. Neither has a waste number, and both are usually worth more than any waste number in the account.

The practical consequence is that a Meta audit should be ranked by how much of the account each finding gates, not by dollars at risk. Signal quality gates everything. Structure gates every ad-set comparison. Creative gates the ceiling. Attribution settings gate whether anyone can tell. A tidy list of small waste items sits below all four, however satisfying it is to report.

Where these thresholds stop applying

• →The 50-events-per-week floor is Meta's documented learning requirement, not a law of nature, and it moves as the platform changes. Treat the arithmetic as the durable part and re-check the constant.
• →Below roughly $10k a month, the ad-set ceiling collapses to one and the honest reading is that structure is not your problem — creative and offer are. Applying a high-spend structural audit at that level produces findings you cannot act on.
• →Fatigue and audience saturation are genuinely indistinguishable in platform data. A decayed asset in a small audience may be a creative problem or an exhausted pool, and the audit can only flag the decay, not diagnose which.
• →Every capacity finding here is unfalsifiable in the short run: you cannot prove signal fixes worked without a holdout, and most accounts will not run one. Be explicit that the expected value is a judgment rather than a measurement, or you are doing the same overclaiming the audit was meant to catch.
• →A clean audit means the account is being fed and read well, which is a precondition for performance rather than evidence of it. The scored definition of good lives separately in the account standard.

The recurring theme: performance is capped by signal and creative, and reported ROAS overstates the truth until reconciled. The structural fixes live in Meta account structure in 2026 and the creative testing system that scales; the measurement fix in MER vs ROAS. Size what a customer is worth with the breakeven ROAS calculator. And for an independent read of your own account, the free audit runs the structural findings without the sales theater.